A genetic feature weighting scheme for pattern recognition

摘要: This paper proposes a new pattern recognition scheme, combining adaptive feature weighting and modified k-Nearest Neighbor (k-NN) rule. The proposed method named adaptive-3FW. It uses three non-uniform weight levels (zero weight, middle full weight) to each feature. value is determined using genetic algorithms (GAs). adaptive-3FW overcomes overfitting issues achieves high performance. Novel GA operators tailored for this formulation are introduced implement the scheme. Further, k-NN which class-dependent strategy. Whilst conventional systems use same set of weights all classes, algorithm different sets classes. Experiments were performed with UCI repository machine learning databases unconstrained handwritten numeral database Concordia University in Canada show performance method.